Stronger Together? Linked Fate and Voter Preferences in the 2020 Election
A. Jordan Nafa, Meredith Walsh Niezgoda, P. DeAnne Roark, and Valerie Martinez-Ebers
University of North Texas
September 15th, 2022
Introduction
-
How do intersecting identities influence candidate preference among members of pan-ethnic racial groups?
- Do these relationships vary within pan-ethnic identity groups?
-
We examine differences within and between pan-ethnic identity groups
- Demonstrate the possibility of moving beyond “Latinos and Asian Americans are not monolithic” and treating heterogeneity among pan-ethnic groups as substantively interesting
Previous Research
Black Americans’ Linked Fate (Dawson 1995)
Latinos’ and Asian Americans’ Complicated Linked Fate (Junn and Masuoka 2008; Sanchez, Masuoka, and Abrams 2019)
Intersectional Linked Fate (Gay, Hochschild, and White 2016; Junn 2017)
Intra-group differences in Linked Fate (Arora, Sadhwani, and Shah 2021)
We attempt to bridge several lines of research on pan-ethnic linked fate that have thus far developed largely independent of one another
Methods and Data
Data
Asian (N = 3,898) and Latino (N = 3,950) Respondents on the 2020 Collaborative Multi-Racial Post-Election Survey (Frasure et al. 2021)
-
Dependent Variable
\(y_{i} = \begin{cases}1 \text{ if respondent } i \text{ Supported Trump in 2020}\\ 0 \text{ Otherwise}\end{cases}\)
-
Dimensions of Linked Fate: Pan-Ethnic, Gender, and Intersectional
- “How much do you think what happens to the following groups here in the U.S. will have something to do with what happens in your life? What happens to [identity group] will have…”
Sex, Citizenship, Age, Education, Ancestry, Partisan Identification, percentage University educated, Median Age
Multilevel Regression and Post-Stratification (MrP)
Bayesian Multilevel Logistic Regression
Partial pooling of categorical predictors to penalize model complexity and prevent over-fitting (Gelman and Hill 2007; Park, Gelman, and Bafumi 2004)
Weakly informative priors on the coefficients and hierarchical variances to stabilize computation (Gelman and Hill 2007; Gelman, Hill, and Vehtari 2021)
For non-census variables we first develop models to predict their distribution in the population and add them to the stratification table
MrP Model Specification
\[ \begin{aligned} y &\sim \mathrm{Binomial}(n, k, \theta)\\ \mathrm{logit}(\theta) &= \alpha + \beta_{1}\text{Female}_{i} + \beta_{2}\text{Citizen}_{i} + \upsilon_{j[i]}^{\mathrm{Ancestry}} + \upsilon_{l[i]}^{\mathrm{Education}} + \upsilon_{m[i]}^{\mathrm{Age}} +\\ &\quad \upsilon_{o[i]}^{\mathrm{Partisan}} + \upsilon_{p[i]}^{\mathrm{Linked~Fate}} + \upsilon_{q[i],p[i]}^{\mathrm{Female \times Linked~Fate}} + \upsilon_{q[i],j[i]}^{\mathrm{Female \times Ancestry}} +\\ &\quad \upsilon_{p[i],j[i]}^{\mathrm{Linked~Fate \times Ancestry}}\\ \text{where}\\ \upsilon_{j}^{\mathrm{Ancestry}} &\sim \mathcal{N}(\left[\gamma_{0} + \gamma_{1}\text{% College}_{j} + \gamma_{2}\text{Median Age}_{j} \right], ~\sigma^{\mathrm{Ancestry}}) \quad \text{for } j \in \{1, 2, \dots, J\}\\ \upsilon_{l}^{\mathrm{Education}} &\sim \mathcal{N}(0, ~\sigma^{\mathrm{Education}}) \quad \text{for } l \in \{1, 2, \dots, L\}\\ \upsilon_{m}^{\mathrm{Age}}&\sim \mathcal{N}(0, ~\sigma^{\mathrm{Age}}) \quad \text{for } m \in \{1, 2, \dots, M\}\\ \upsilon_{o}^{\mathrm{Partisan}} &\sim \mathcal{N}(0, ~\sigma^{\mathrm{Partisan}}) \quad \text{for } o \in \{1, 2, \dots, O\}\\ \upsilon_{p}^{\mathrm{Linked~Fate}} &\sim \mathcal{N}(0, ~\sigma^{\mathrm{Linked~Fate}})\quad \text{for } p \in \{1, 2, \dots, P\}\\ \upsilon_{q,p}^{\mathrm{Female \times Linked~Fate}} &\sim \mathcal{N}(0, ~\sigma^{\mathrm{Female \times Linked~Fate}}) \quad \text{for } q \in \{1, 2, \dots, Q\} \text{ and } p \in \{1, 2, \dots, P\}\\ \upsilon_{q,j}^{\mathrm{Female \times Ancestry}} &\sim \mathcal{N}(0, ~\sigma^{\mathrm{Female \times Ancestry}}) \quad \text{for } q \in \{1, 2\} \text{ and } j \in \{1, 2, \dots, J\}\\ \upsilon_{p,j}^{\mathrm{Linked~Fate \times Ancestry}}&\sim \mathcal{N}(0, ~\sigma^{\mathrm{Linked~Fate \times Ancestry}}) \quad \text{for } p \in \{1, 2, \dots, P\} \text{ and } j \in \{1, 2, \dots, J\} \end{aligned} \]
Post-Stratification Stage
Post-Stratified estimate by Linked Fate and Ancestry or Sex based on the 2019 American Community Survey 5-year Integrated Public Use Microdata Series data (Ruggles et al. 2022)
\[\begin{aligned}\theta^{\mathrm{MrP}} = \frac{\sum_{p,j \in P,J}N_{p,j}\cdot \theta_{p,j}}{\sum_{p,j \in P,J}N_{p,j}}\end{aligned}\]
\(\theta\) is the predicted probability of support for Trump from our model
\(N\) is the observed cell count in the post-stratification table
Summing over the population predictions within each pair of cells \(p,j \in P,J\) and dividing by the population total yields \(\theta^{\mathrm{MrP}}\) which represents the census-adjusted parameter estimate
Results
Social identities in the Era of Trump
Race, ethnicity, gender, and social identity theory
What makes some identities more salient than others?
The Trump presidency and the salience of group identity
Theoretical Expectations
The salience of both pan-ethnic and gender-based dimensions of linked fate will vary across different sub-groups
The threat against multiple identities will activate intersectional linked fate
Relationship between intersectional linked fate and candidate preference is greater than the gender or pan-ethnic dimensions alone